PART II: EMPIRICAL RESULTS
A. Attrition 112
4. Evaluation
4.1 Identification and Methodology 124
Amount (US$) Starting Date
Finalizing Date 1. Revitalization of the City Center 14,098,000 Mar-09 Jun-13
Obra Orla Ferroviária 2,600,000 Feb-11 Apr-13
Orla Morena 1st Stage 6,498,000 Mar-09 Dic-10
Orla Morena 2nd Stage 5,000,000 Feb-11 Jun-13
2. Transport and Mobility 20,906,000 Nov-09 Jul-13
Via Morena 10,071,000 Nov-09 Dic-12
Avenida Júlio de Castilho 10,835,000 Ago-11 Jul-13 Infrastructure Work
125
Given the retrospective character of the evaluation and the availability data, I will use a differences in differences approach. This methodology compares the changes in outcomes over time between a population that is enrolled in a program (the treatment group) and a population that is not (the comparison group). By comparing the enrolled before and after, we are controlling for factors that are constants over time (since we are comparing the same group to itself). And, by comparing the enrolled group with a group that is exposed to the same conditions, we are controlling for time-varying factors. In this way we are controlling for the most worrisome sources of bias (self-selection and time-varying factors).
With the available data, the smallest unit of assignment available is the property level, so that will be the unit of analysis. To determine which properties belong to the treatment group and to the control group I will use the criterion of geographical proximity to the interventions. Thus, I will assume that those properties that are closest to the interventions are the ones that will benefit from the investment and, therefore, will belong to the treatment group, whereas those that are further away from the works will belong to the control group. To determine the location of the property in respect to the infrastructure works I will use the most accurate possible measures.
Since there are no GPS coordinates for the properties, the smallest geographical unit that allows us to allocate accurately the property is the variable “neighborhood”. Therefore, the properties belonging to adjacent neighborhoods of the interventions will be considered the treatment group, and the properties of the more remote areas will be used as a comparison group.
Although the properties that are located in different areas of the municipalities may have different values (e.g. property in the center versus properties on the periphery), all properties are simultaneously exposed to economic conditions within the municipality, so we would expect that
126
changes or macroeconomic "shocks“ on municipal properties will affect both, the treatment and control. Therefore, the control group will control for time variations.
There is also information on the dates when the interventions started and finalized. I will assume that individuals anticipate the change in the values as soon as the works start to be implemented, therefore, I will consider the starting date of the works as the beginning of the treatment period.
Finally, since the objective of the program is to improve the quality of life of the population of the municipality and contribute to economic competitiveness, in my analysis I will discard territorial properties, as these are in outlying areas of the city center and selling price is governed by different criteria from the rest of the properties. Even more, to ensure comparability of the observations in the sample, the analysis will focus on residential properties.
Hedonic Prices
To quantify the benefits of the infrastructure works I will use the hedonic price approach using the variation in housing prices. The logic behind that is that, if the interventions have improved quality of life of the citizens, the desirability in terms of demand will be reflected in an increase in housing prices in the neighborhood. Thus, the outcome indicator that I use to conduct the impact analysis is the price of real estate (houses, buildings).
According to the hedonic price model (Griliches, 1979) the price of a good is determined by the implicit price of each of its components. In this case, the price of real estate would be formed by the implicit prices of attributes such as number of rooms, quality of materials and provision of urban infrastructure services, among others. In a competitive market, price is determined by the equilibrium in which the functions of demand and supply of buyers and sellers are equal.
127
According to the theory of hedonic prices, changes in real estate prices by varying one of its attributes (and keeping everything else constant) determine the valuation of individuals of that attribute. In our case, the change in housing prices by providing them with new urban infrastructure reflects what must be paid to the individual to maintain their standard of living.
The marginal willingness to pay for each of the attributes available can be used to infer the welfare effects of a marginal change in one of the attributes for individuals.
In this context, it may be concluded that the intervention of urban infrastructure has a positive impact if the price of housing in the treatment group (those who have benefited from urban investments) is greater than it had been in the absence of investment (estimated by the control group or those of similar homes that have not been beneficiaries of investments).
Importantly, there are some limitations when using prices as an indicator of impact. Specifically we are building on the assumption that markets work well, however, in some countries or regions may dominate certain degree of informality or allocation of land and buildings that are not commercially available or can be barriers to mobility that make prices do not collect welfare. On the other hand, there are advantages to study property prices versus other variables. For example, prices account for the effects faster than other outcome variables such as reallocation of businesses, employment rates, etc... Also, in areas with certain degree of development, data on real estate prices are typically available in administrative records of fiscal agencies or agencies buying and selling real estate.
128 Identification
Figure 1 shows a map of the area of interventions and the name of the nearby neighborhoods.
The names of the neighborhoods that are circled are the ones that are assigned to the treatment group. Those of the left-hand part of the map circled in black and red belong to the transportation component; whereas those at the right-hand side of the map circled in green belong to the revitalization component.
Figure 1. Map of the Investments of Procidades in Campo Grande.
Table 2 shows when the components of the analysis start and finalize (details are specified in Table 1).
Table 2. Timeline of interventions of the urban revitalization and transportation of Procidades in Campo Grande.
129
Specification
The following model is specified to identify impact:
𝑃𝑖𝑠𝑡 = 𝛼𝑇𝑠𝑡+ 𝜇𝑠+ 𝜋𝑡+ 𝛽𝑋𝑖𝑠𝑡+ 𝛾𝐾𝑠+ 𝜀𝑖𝑠𝑡 (1)
Where 𝑃𝑖𝑠𝑡 is the logarithm of price per square meter of property i located in neighborhood s in semester t, Tst is a variable that takes the value 1 for the treatment neighborhood from the time when begins treatment and 0 otherwise, μs is a dummy variable per neighborhood, πt is a dummy variable for period of time t, Xist are observable characteristics of the property, 𝐾𝑠 is a dummy variable that equals 1 for neighborhoods with other interventions of urban infrastructure and 𝜀𝑖𝑠𝑡 is the error term, that contains the unobservable characteristics of the property price. Standard errors are clustered at the neighborhood level. I will assume that the property price adjusts instantaneously to changes in the expected value. The coefficient α captures the aggregate impact of the intervention from the time of the start of the intervention.
We also analyze the differential effects before, during or after the period of implementation of the works. For this model 2 is estimated:
𝑃𝑖𝑠𝑡 = ∑𝑗=𝑛𝑗=1 𝛼𝑗𝑇 𝑠𝑡𝑗 + 𝜇𝑠+ 𝜋𝑡+ 𝛽𝑋𝑖𝑠𝑡+ 𝛾𝐾𝑠+ 𝜀𝑖𝑠𝑡 (2) Starting
Date
Finalizing Date 1. Revitalization of the City Center Mar-09 Jun-13
2. Transport and Mobility Nov-09 Jul-13
Infrastructure Work
130
Where 𝑇 𝑠𝑡𝑗 is a dummy variable that equals 1 for treatment neighborhoods during the treatment and 0 in the remaining periods of time. The other variables are interpreted as in equation (1).
5. Data
The main sources of information about property prices come from two administrative databases managed by the municipality:
(i) “Tax on Transmission of Real Property” database (ITBI for its Portuguese acronym55) contains information on property prices of transactions in real property. It includes variables such as date of registration of the property, type of transaction, property prices and payment date of the transaction, and variable location of housing (including address, area, district, area, subdividing).
(ii) Urban Building and Land Tax (IPTU for its Portuguese acronym56) database contains historical data on the basic characteristics of the properties such as area, number of rooms, water service provision, paving, lighting, telephone, urban, materials of the walls, floor, ceiling, roof, and other characteristics of the construction.
There may be other investments in the municipality during the time of the intervention that may be biasing our estimates. For instance, if they are constructing urban infrastructure assets in a
55 Imposto sobre a transmissão de bens imoveis
56 Imposto Predial e Territorial Urbano
131
neighborhood that belongs to the treatment group, we will be counting it as part of our program and therefore we will be overestimating the impact. On the other hand, if they are constructing in the control group we may be underestimating the impact. Therefore, data on other interventions during the period of analysis were also used to control for those bias.
According to ITBI 21,355 properties were sold between February 2008 and November 2013.
Table 3 shows its distribution along the different kind of properties. For the reasons mentioned above, we will focus in residential properties.
Table 3 Frequency Chart type of properties sold in Campo Grande (2008-2013)
Source: administrative data ITBI Campo Grande
The database of IPTU has data on all the properties in Campo Grande from the years 2005 to 2013. The database contains information on basic characteristics such as area, water utilities, paving, lighting, telephone, urban planning, and other.
Figure 2 below shows other interventions conducted in the municipality besides the program that we want to estimate and the neighborhoods that are affected.
USOIMOVEL Freq. Percent Cum.
COMERCIAL 110 0.52 0.52
FINALID ESSENCIAIS 2 0.01 0.52
INDUSTRIAL 2 0.01 0.53
MISTO 105 0.49 1.03
RELIGIOSO 6 0.03 1.05
RESIDENCIAL 12,634 59.16 60.22
SERVICOS 309 1.45 61.66
TERRITORIAL 8,187 38.34 100
Total 21,355 100
132
Figure 2. Map of all the urban infrastructure interventions carried out in the municipality of Campo Grande between 2008 and 201257
Therefore, the database used for the analysis of Campo Grande consists of 12,634 observations of the residential properties sold in Campo Grande in the available period February 2008 to November 2013. For each residential property sold, we have the price, date of sale, neighborhood, basic observable characteristics, and on the other hand we count with information on other interventions in the municipality.
Table 4 shows the distribution of observations in the treatment group and control throughout the semester for the overall program and for each of its components.
57 Notice we don’t have data of new interventions started in 2013.
133
Table 4. Sample distribution by semester for treatment and control groups for different interventions
A. Overall Program
B. Revitalization* C. Transport and Mobility
*Includes 110 commercial properties that will be used in the final estimations of the revitalization component only.
Fuente: Municipal administrative records from IPTU and ITBI
Periodo Control Tratamiento Total
134
6. Results
In the following sections we study the impact of the program as a whole and separating the revitalization and transportation components.
Graphic Analysis
Figure 3 shows the evolution of the variable of interest (the logarithm of price per square meter) for the treatment group and the control group. The results are adjusted for observable characteristics of the properties and control various other program interventions have been carried out in the town. The vertical lines show the beginning and end of the period of implementation of the interventions. The horizontal axis includes the periods of time being 1 the first semester 2008, 2 the second semester of 2008 and so on until the period 12 corresponding to the second semester of 2013.
According to the graphs, it seems that the pre-intervention period evolves similarly in all the situations. Although the equality of trends needs will be tested more rigorously, it provides credibility to the comparability of treatment and control groups defined in the study.
Figure 3. Evolution of semi log price per square meter of property1 set in the town of Campo Grande between 2008 and 2013
135
A. Overall Intervention
B. Intervention of Revitalization C. Intervention of Transportation
1Adjusted by observable characteristics of the properties and for other interventions conducted in the municipality.
Source: administrative data from ITBI, ITPU and information of other works in the municipality of Campo Grande.
Pre-intervention equal trends test
This section formally analyzes the pre-program treatment group and the comparison group trends. Table 5 shows the p-value for the results of a joint significance F test that compares the value of the slopes between the treatment group and the comparison group in the pre-intervention period.
136 Table 5. Joint significance F-test of equal pre-trends
Interventions p-value of the F-test of equal trend for the pre-intervention periods 1
Overall Program 0.9402
Intervention of Revitalization 0.7958
Intervention of Transportation 0.6623
1Adjusted by observable characteristics of the properties and for other interventions conducted in the municipality.
Source: administrative data from ITBI, ITPU and information of other works in the municipality of Campo Grande
As we see, in all cases we cannot reject the null hypothesis that the trend before the intervention is equal to 95% probability.
Impact of the program
Table 6 shows the result of equations (1) and (2) for different types of interventions.
Column 1 shows the result of equation (1) for the overall intervention. In column 2 the effect is decomposed by period of time. According to these results we would conclude that there are no significant impacts of the intervention (only a positive and significant impact of the program is observed in the first half of 2011.)
Models 3 and 4 show the analogous results but for the revitalization component. As shown, it seems that there is a significant negative impact of 5.3%. Decomposing the effect by periods we see that the negative impact comes from the period the second half of 2012 and first half 2013.
Since this component is concentrated in the city center and these areas usually present special characteristics, we explore this result in more detail below.
137
Finally, columns 5 and 6 show the results for the interventions on transportation. The average impact of these interventions is 6.7% and significant at the 95% level. Decomposing the effect by semesters, we see that there is a negative (although non-significant) impact at the beginning.
It could be explained by the inconveniences of the construction works that could have been offsetting the anticipation of the positive effects of the new roads. At the first half of 2011 the intervention starts to show a positive and significant effect that lasts until the second half of 2013. According to the time of implementation of works this increments match with the construction phase of the Avenida Julio Castilho, suggesting that the construction of this road had an immediate impact on prices of nearby properties.
Table 6. Impact of Procidades in the logarithm of the prices by squared meter in the Municipality in Campo Grande (2009-2013)1
138
Muestra: Todo el municipio Yes Yes Yes Yes Yes Yes
Muestra Solo region Centro No No No No No No
*** p<0.01, ** p<0.05, * p<0.1
1NOTES:
1. Inlcudes fixed effects at the sector level, at the period level, and controls for the characteristics of the propoerties and neighborhoods affected by other interventions. The property controls include: whether it is an appartment, the area of the lot, the area of the swimming pool, whether it was constructed before 2000, whether it has access to public transportation, whether it has access to municipal cleaning servicies, whether it has any of the following: water, garbage service, sewage, illumination, curb, paving, electricity, telephone, sidewalks. Also by the characteristics of the materials of the interior and exterior finish of the walls of the building, the roof, ceiling, window frames, structure of the building, floor, installation of electrical and sanitary installation, state of preservation, whether there is a lift and if it is in a regular or irregular situation.
Overall Program Intervention of Revitalization
Intervention of Transportation
Robust standard erros in parenthesis
2. Requalification Interventions include Orla Ferroviaria, Orla Morena, and Transportation Interventions includes Via Morena and Julio de Castilho.
1. Inlcudes fixed effects at the neighborhood level, at the period level, and controls for the characteristics of the propoerties and neighborhoods affected by other interventions. The property controls include: whether it is an appartment, the area of the lot, the area of the swimming pool, whether it was constructed before 2000, whether it has access to public transportation, whether it has access to municipal cleaning servicies, whether it has any of the following: water, garbage service, sewage, illumination, curb, paving, electricity, telephone, sidewalks. Also by the characteristics of the materials of the interior and exterior finish of the walls of the building, the roof, ceiling, window frames, structure of the building, floor, installation of electrical and sanitary installation, state of preservation, whether there is a lift and if it is in a regular or irregular situation.
2. Requalification Interventions include Orla Ferroviaria, Orla Morena, and Transportation Interventions includes Via Morena and Julio de Castilho.
139
Alternative Estimates of the Impact of the Revitalization Component
Revitalization interventions are concentrated in the city center, the historic area of the municipality. Due to the special and unique characteristics presented in the city center and the bordering areas, it is especially difficult to find comparable neighborhoods. Thus, in this section I explore alternative definitions of the comparison group.
Other limitation of this part of the study is that there are a relatively small number of residential properties sold (see table 7). To try to increase the number of observations, I take a broader definition of properties and include properties dedicated to commerce and services. However, only 110 of these property sales were recorded during the study period, so the results do not change much. Columns 1 and 2 of table 6 show the result of the estimation of the new sample.
Once this is done, we address the limitation of the particularities of the city center. The main limitation that city centers present is that it usually has unique features that make it difficult to compare with other areas of the cities. In particular, if the neighborhoods of the municipality included in the comparison group have growth rates that are not comparable with the city center in the absence of the program, our estimation could be biased. To control for this potential bias, I selected as controls only the sub-sample of neighborhoods adjacent to the treatment areas, but where no intervention was performed. Columns 3 and 4 of Table 7 show the result of this regression. As we see, using the new comparison group no significant differences were found between the treatment group and control58. One possible explanation for this could be that given the geographical proximity of the neighborhoods there is no effect because the control group is also benefiting from the impact of the construction of urban upgrading interventions.
58 The p-value of the pre-treatment equal test is 0.1659, so we cannot reject equal trends at an 80% of probability.
140
Table 7. Estimating the impact of interventions Revitalization component in the log prices per square meter of properties in Campo Grande (2008-2013)1
(1) (2) (3) (4)
Incluye inmuebles comerciales y de servicios Si Si Si Si
Incluye inmuebles comerciales y de servicios Si Si Si Si